VesperMUSE is a multi‑modal vision transformer that enhances image semantic segmentation under varied lighting by combining physics‑informed low‑light adaptation, thermal‑guided low‑light image enhancement, and surface normal maps. The framework fuses RGB, thermal, and normal features through a modality‑agnostic self‑attention mechanism, enabling robust segmentation without architectural changes for different sensor setups. Extensive tests show that VesperMUSE recovers missing visual cues and achieves state‑of‑the‑art performance in extreme, unstructured environments.
By Raul Balmez
MMSearch-OPD is a new approach for long‑horizon multimodal search agents that combines a controllable data refinement pipeline with joint reinforcement learning and on‑policy distillation. The refinement process expands reasoning chains from existing simple multimodal questions, allowing adjustable search depth and cross‑modal evidence requirements while reusing image–question pairs. Training uses answer rewards, behavior penalties, and a decaying distillation coefficient to guide the agent, and experiments demonstrate that both data refinement and joint training after supervised fine‑tuning improve long‑horizon multimodal search performance.
By Huanyao Zhang, Jiepeng Zhou, Jiaye Lin, Ontai Tang, Jingtian Hao, Mugeng Liu, Ru Peng, Lang Mei, Xiaohan Yu, Liyan Liu, Chong Chen, Wentao Zhang
MJev-Spatial is a multimodal decision model that unifies spatial intelligence tasks—choice, numerical, and point prediction—into a single classification framework. It treats each candidate as a class, scoring them with pretrained query and key projections without adding parameters, and achieves predictions in a single forward pass with no answer tokens. Trained on 12.3 M question‑answer pairs, it outperforms eight baseline models, ranks first on seven of twelve benchmarks, and runs 2.5× faster than the fastest autoregressive baseline.
By Luozheng Qin, Yuhan Zhu, Changhai Zhou, Qian Qiao, Jun Gao, Yuhua Zhou, Harry Zhang, Xiangyu Zeng, Yuxuan Liu, Kaijie Chen, Andrew Chen, Pony Ma
EPIC-Bench is a perception‑centric benchmark for fine‑grained embodied visual grounding in vision‑language models. It contains 6.6k human‑annotated tuples of Image, Text, and Mask across the three core stages of embodied interaction—Target Localization, Navigation, and Manipulation. The benchmark offers 23 fine‑grained tasks, multi‑dimensional settings, and evaluates over 80 leading VLMs to provide diagnostic insights into their vision‑language grounding capabilities.
By Haozhe Shan, Xiancong Ren, Han Dong, Yingji Zhang, Haoyuan Shi, Jiayu Hu, Qinfan Zhang, Yi Zhang, Yong Dai, Bin Shen, Lizhen Qu, Zenglin Xu, Xiaozhu Ju
GRADS is a graph-guided framework for generating synthetic data that explicitly controls difficulty by maintaining a graph of evidence and inference dependencies. It iteratively selects seeds, expands evidence, constructs question–answer pairs, and verifies them, ensuring that each generated task contains only indispensable inference steps. Experiments across security, software engineering, and scientific domains show that GRADS produces verified tasks with distinct difficulty levels, and training models on this data yields significant performance gains, including high solution-termination rates and reduced tool usage.
By Gaocheng Yang, Ruijie Tao, Kai Ye, Senhua Zhang, Zhengqin Zhu, Ziyang Li
LEASH is a training‑free framework that enables uncertainty‑aware escalation from a weak to a strong computer‑use agent. It uses latent uncertainty signals from the weak agent’s hidden states to decide when to hand off control, and a shadow observer to maintain oversight of the strong black‑box agent. Experiments on OSWorld show LEASH improves task success over standalone weak agents and outperforms random escalation, achieving comparable performance to Claude Opus 5 while cutting inference cost by 19.2%.
By Qingni Wang, Changdae Oh, Yuzhe Yang, Sharon Li, Jiachen Yang, Ang Li, Xin Eric Wang
SCARCE (Shared Compute ARena for Coding agEnts) is a framework where autonomous agents tackle coding tasks with individual deadlines while negotiating shared resource access. The study observes that agents spontaneously form informal institutions—rules and commitments based on reputation, reciprocity, and precedent—that persist across tasks and influence resource allocation. A 21‑category taxonomy of negotiation strategies and emergent norms was developed to analyze eight models, revealing that newer frontier models exhibit more institutional behaviors, alongside exclusionary coalitions and rare divide‑and‑conquer tactics.
By Abhinav Kumar, Rediet Abebe, Eugene Bagdasarian, Sahar Abdelnabi
OS-Helix is an end‑to‑end multimodal agent that unifies graphical user interface (GUI) actions and code execution into a single action space. It is trained on high‑quality hybrid data generated through application‑artifact composition and a decoupled instruction/evaluator pipeline, and further refined with online reinforcement learning using a step‑balanced GSPO algorithm. The resulting models achieve significant performance gains over hybrid baselines on the OSWorld‑Verified benchmark and demonstrate cross‑platform transfer to WindowsAgentArena.
By Bowen Yang, Kaiming Jin, Zichen Ding, Zhenyu Wu, Zehao Li, Zhoumianze Liu, Kanzhi Cheng, Fangzhi Xu, Yanan Sun, Qiushi Sun, Yu Qiao
The paper introduces a domain‑specific debugging agent for repairing near‑miss accelerator kernels, arguing that debugging is more efficient than regenerating kernels from scratch. It tackles knowledge scarcity, integrity, and cost control through pattern retrieval, anti‑cheat detection, and convergence safeguards. Empirical results show the debug agent achieves a 66.7% Pass@1 rate and uses 92.8% fewer tokens per success compared to traditional regeneration methods.
By Yansong Sun, Shenxiu Wu, Siyuan Chen, Runlin Hou, Junhao Qiu, Junming Cao, Shudi Shao, Zhichao Lu, Qingfu Zhang
TimeRSI is a self‑improving deployment system for time‑series foundation models that automates the creation and refinement of forecasting pipelines. A coding agent translates task descriptions into executable changes, tests each revision on historical data, and uses error feedback to guide further revisions. During live deployment, an adaptive selector chooses the best pipeline based on completed forecasts under similar conditions, achieving superior performance on 198 tasks compared to all other submissions.
By Haoxin Liu, Deqing Fu, Lucheng Fu, Yiqiao Jin, Rushi Qiang, Harshavardhan Kamarthi, B. Aditya Prakash
DeepUnfolding Agent is a schema‑driven FastAPI platform that exposes deep‑unfolded adaptive control as an external software service, separating controller logic from domain‑specific execution. It allows external executors to register strategy schemas, submit diagnostic variables, receive bounded control actions, execute optimization rounds, and return feedback for controller updates. The framework includes a strategy registry, diagnostic‑to‑state adapter, controller factory, safety guard, executor abstraction, feedback update path, SQLite persistence, Docker deployment, and reproducible benchmark scripts, supporting classical, quantum‑aware, hybrid, Qiskit Aer, and IBM quantum processing pathways.
By Shanika Iroshi Nanayakkara, Shiva Raj Pokhrel
The paper presents a self‑improving agent that expands its own domain‑specific language (DSL) without external reward, judge, or answer key. Using two nested loops, the agent only accepts new operators when two independent implementations agree and a conservation law holds, thereby ensuring exact correctness. Experiments show the agent triples its answerable‑question set on a transaction table and extends its capabilities across unrelated domains, outperforming a no‑induction baseline.
By Yun-Shiuan Chuang, Yirou Ge, Yixi Li, Yiming Bian, Zijie Pan, Jenna L. Ballard, Uma Kona, Linsey Pang, Prakhar Mehrotra
AgentEval is a system that tracks a contract-bound assessment record throughout a stateful agent’s tool execution, capturing state, events, and evidence to determine whether obligations are satisfied, violated, or unresolved. In 32 tool‑world administrations, independent checks confirmed all 30 complete episodes, while controlled interventions on 48 evidence views uncovered 36 binding faults that hash‑only checks missed. A public‑benchmark study on 192 held‑out executions identified ten completed subgoals within broader task failures and highlighted the impact of missing evidence on overall task success.
By Ruiqi Gao (Alibaba Cloud, Alibaba Group)
All-in-One Memory (AiO-Mem) is a reinforcement learning framework that unifies memory retrieval, writing, and stopping decisions into a single reward space. It compiles diverse histories into traceable multi‑view events, uses a unified policy actor to generate interaction trajectories, and employs a trajectory‑conditioned process reward model trained with supervised fine‑tuning and DPO. Experiments on long‑context memory benchmarks show AiO‑Mem consistently improves answer accuracy compared to static memory systems and other RL‑trained agents.
By Bingbing Wang, Jingjie Lin, Weiming Qiao, Jing Li, Ruifeng Xu
Diviner is an agentic root‑cause analysis system designed for large‑scale distributed infrastructures. It organizes incident investigation into a four‑stage reliability lifecycle, leveraging a corpus of reviewed incidents and service‑owner expertise to create scoped, versioned skills, workflows, typed tools, and an incident knowledge graph. The system employs adaptive lead and context‑isolated workers to apply relevant knowledge to real‑time evidence, evaluates investigation trajectories against responder‑reviewed outcomes, and updates operational knowledge through feedback loops that require replay, quality checks, and owner approval.
By Pradipta Ghosh, Gaurav Mitra, Shripad Nadgowda, Akshil Shah, Junyu Liu, Canturk Isci, Syamla Bandla, Xinyao Hu, Onkar Deshpande, Anik Dasgupta
Mallea is a system that builds, steers, and memorizes worlds from text by using a coding agent that infers object interaction requirements and plans object structure, properties, and rules for compact representations. The world evolves under these rules, with optional user actions steering it and text requests editing properties or extending capabilities while preserving unaffected state. Object texture memory retains reusable appearance for whitebox-defined objects and adapts it as those objects change, while a generative renderer combines current structure with remembered appearance to advance the world state.
By Xuanyi Liu, Ruiqi Li, Siwei Ma
The paper introduces a test-time aggregation framework for multi-agent reasoning with large language models. It mitigates non-semantic biases by normalizing reasoning traces and injects bias to create decision instability, which signals reliability. The method improves accuracy and token efficiency on several mathematical reasoning benchmarks without requiring validation data or prior model strength knowledge.
By Sungjae Lee, Soheil Abbasloo, Jabin Koo, Jungseul Ok
ARGUS is a neuro‑symbolic, agenda‑driven framework designed to improve the reliability of user simulators in interactive environments. It separates state tracking from natural language generation by using a dynamic agenda stack and symbolic operations to enforce goal progression and constraint‑aligned interactions. The authors distill the model into compact versions via evaluator‑guided SFT and online RL, and evaluate its behavioral fidelity with a new Composite Behavioral Score metric on $ au^3$‑Bench and MultiWOZ, showing significant performance gains.
By Hang Zheng, Nan Jiang, Zichen Zhu, Zixu Yang, Zhiyang Tang, Shuai Fan, Lu Chen, Xiaobao Wu, Kai Yu
Jev-LiteMAS is a multi‑agent framework that separates decision‑making from generative execution in large‑language‑model workflows. By using a lightweight decision model (Jev) to dynamically select experts and screen artifacts, it closes the planning and review gaps left by pure generative planners and reviewers. Experiments on four benchmarks show that Jev‑LiteMAS cuts generative‑LLM token usage by 12.7–54.6% while keeping accuracy within 1.34 percentage points and reducing latency in most settings.
By Kaixiang Wang, Zhaijiacheng Zhou, Jiong Lou, Jie Li
The paper introduces TIR-EC, a tool‑integrated reinforcement learning framework designed to improve large language models (LLMs) for emergency department prediction tasks. TIR-EC exposes groups of electronic health record (EHR) features as callable tools and trains the LLM using rewards from a gradient‑boosted tree teacher, while a reverse curriculum gradually removes prompt‑based skills to embed task‑specific clinical knowledge into the model weights. Experiments on three clinical prediction tasks show that a 4B TIR‑EC agent achieves the highest macro F1 among compared models, surpasses conventional machine‑learning baselines, and competes with much larger frontier LLMs, while also selecting more useful evidence with fewer tool calls.
By Pengfei Hu, Chenwei Wu, Qiao Jin, Jun Deng, Zhiyong Lu, Liyue Shen, Yue Ning